An Image Analysis for the Development of a Skin Change-Based AI Screening Model as an Alternative to the Bite Pressure Test

Abstract

Background/Objectives: Oral function assessments in hospitals and nursing facilities are mainly performed by nurses and caregivers but are sometimes not properly assessed. As a result, elderly people are not provided with meals appropriate for their masticatory function, increasing the risk of aspiration and other complications. In the present study, we aimed to examine image analysis conditions in order to create an AI model that can easily and objectively screen masticatory function based on occlusal pressure. Methods: Sampling was conducted at the Hokkaido University of Health Sciences (Hokkaido, Japan) and the university’s affiliated dental clinic in Hokkaido. Results: We collected 241 waveform images of changes in skin shape during chewing over a 20 s test period from 110 participants. Our study used two approaches for image analysis: convolutional neural networks (CNNs) and transfer learning. In the transfer learning analysis, MobileNetV2 and Xception achieved the highest classification accuracy (validation accuracy: 0.673). Conclusions: Therefore, it was determined that analyses of waveform images of changes in skin shape may contribute to the development of a skin change-based screening model as an alternative to the bite pressure test.